(474f) Uni-MOF: A Universal Material Representationlearning Framework for Metal-Organic Frameworks

Gas separation process is ubiquitous in industrial production, compared with conventional separation technologies, advanced nanoporous materials such as Metal-Organic Frameworks (MOFs) have advantages in terms of energy and separation efficiency due to their tunable pore structure and chemical affinities. The crucial issue is how to rapidly predict the gas adsorption capacity of MOF adsorbents, however, time-consuming synthesis and adsorption experiments as well as expensive simulations are not preferred. Therefore, we build a universal material learning framework based on transformer for MOFs (Uni-MOF). This pre-trained representation learning framework extracts structural information from hundreds of thousands of 3D materials with no labels, which enables precise prediction of gas adsorption amount based on small data sets and cross-system prediction scenarios. This pre-trained model is also versatile for screening MOF materials under different application criteria.